Odd behavior of a "SAS-2" backplane with SGPIO commands

[Rich <rercola@xxxxxxxxxxx>]

[My apologies if this is the wrong mailing list for this, but it
seemed to be the most relevant. Please direct me elsewhere and discard
if this is OT.]

Hi all,
I've been beating my head against SGPIO for a bit now, and I can't
tell if I'm just misreading SFF-8485, or this backplane is just not
doing something sane.

I have here a Supermicro X8DAH-F+ with a backplane of model
BPN-SAS-836A - which means it has 3 MG9072 controller chips on it.
Supports speaking in-band or out-of-band to read data, control lights,
&c.

With two LSI 9201-16i HBA attached, manipulating it via
/dev/bsg/sas_hostN and smp_utils 0.97, SFF-8485 section 8.4.4 seems to
think that changing this:
 00     41 02 00 00 a0 a0 a0 a0  a0 a0 a0 a0 a0 a0 a0 a0
 10     a0 a0 a0 a0 00 00 00 00  00 00 00 00 00 00 00 00
 20     00 00 00 00

By doing this:
# smp_write_gpio -d a1,a0,a0,a0 -t 3 -i 0 -c /dev/bsg/sas_hostN

Resulting in:
 00     41 02 00 00 a1 a0 a0 a0  a0 a0 a0 a0 a0 a0 a0 a0
 10     a0 a0 a0 a0 00 00 00 00  00 00 00 00 00 00 00 00
 20     00 00 00 00

Should enable a single drive's fault LED (drive m+3, by spec above).

In practice, I am in possession of ~90 identical machines, and on
every single one, this controls the fault LED on 3 drives - I have 3
SAS ports wired to the backplane (via iPASS cable) per HBA, so I would
suspect the HBA of setting bit m+3 on each SAS port [and having each
of its three ports running to an input controlled by a different SGPIO
controller on the backplane]. No permutation of the register bits
beyond the 4 bytes at index 0 changes the behavior of the LEDs that I
can see, and the output in the beginning [a0 a0 ... 00 00 00 00] is
what is returned on clean cold boot of the machine.

In contrast, when I have an LSI 9240-8i attached, I can control
per-drive LEDs, though the mechanism it uses for this is opaque to me
(the megaraid_sas driver doesn't expose a virtual SAS host port like
mpt2sas, so I am manipulating this using LSI's closed-source MegaCli
blob); so I know it is possible, through some method, to control
per-drive LEDs.

I cannot tell, however, whether I am using smp_write_gpio incorrectly
(either by misreading the spec or somehow abusing the virtual SAS
port), or it's a bug somewhere (be it in smp_write_gpio, the kernel,
the HBA firmware, or the upstream MG9072's firmware, that is somehow
magically avoided by the 9240-8i), and there is almost no public
documentation of people using smp_write_gpio.

I've tried this with kernel 2.6.32-220.7.1.el6, 3.4.4, and 3.5.0-rc5
(the last because it includes mpt2sas 13.XX.XX.XX, while 3.4.4 has
12.XX.XX.XX, and I was curious if anything had changed), and smp_utils
0.97.

I wanted to see what the register state was when the 9240-8i was in
use, but I couldn't coax the megaraid_sas driver into disgorging this
information, and upon unplugging the iPASS cable from the backplane to
connect it to the HBA, the backplane appears to reset its internal
state; the LEDs cease blinking in any pattern, and consequent querying
with the HBA returns expected default initial state.

Am I insane, or is this backplane crazy? Does anyone have any
experience with cajoling such things into submission?

Thanks,
- Rich Ercolani
--
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html